>Due to several complaints of unacceptable response time when restoring
>hard drive contents after catastrophic failures, we are considering
>enabling collocation on our VM ADSM server.  I'm looking for comments
>from other sites that have switched from non-collocated storage to
>collocated storage.
>
>1) What kind of increase in tape mounts per nightly backup job are we
>looking at?
>
  Our configuration consists of 3G of 3380 disk (BACKUPPOOL) and 3480
cartridge tapes (BACKUPPOOL1). On a migration process (98% to 70%), which
usually occurrs about twice a week, it used to require about 6 mounts to
migrate about 1G of data, now it requires about 20 mounts to move the same
amount of data. I have also discovered that the number of mounts increases
exponentially when you lower the Low Migration threshold (i.e  changing from
70% to 60% would increase the number of mounts to about 40).

>2) We already have several hundred 3480 tapes worth of saved user data, many
>of which are nearly half empty.  I understand from the documentation that
>enabling collocation will essentially add at least an extra tape for each
>node to our usage.  What are the best methods for avoiding a situation in
>which we have a large set of nearly-empty tapes created after collocation
>as well as our existing set?  Should I define a new storage group with
>collocation enabled and then migrate all data from our current storage
>group it it?
>
 What I did, was to create a list of the tapes in BACKUPPOOL1 and then after
I updated the storage pool to turn collocation on, and I set the Reclamation
Threshold to 100% (this kept the list static).  I went through the list and
did a MOVE DATA vvvvvv STG=BACKUPPOOL for each tape on the list. This would
force a migration about twice a day and the data written to the free volumes
in BACKUPPOOL1 was collocated. It took about a month to convert the pool of
about 200 tapes to a colocated pool of 400+ tapes. This method appeared to
be preferable to moving from tape since each of the input tapes had up to 60
collocation clusters (nodes) which would have required that many output tape
mounts.

 I currently have about 600 tapes in BACKUPPOOL1 with about 56% utilization
(the reclamation threshold is currently at 40%). The alternative to
collocation in our environment would be totally unacceptable. Potentially,
it could have required 100 or more tape mounts to restore a full hard disk.
With collocation, the average work station now only requires 1 or 2 tape
mounts to restore the disk.


 -----------------------------------------------------------
   Don Coleman,
   Queen's University Computing & Communications Services
   Kingston, Ont. Canada  K7L 3N6
   Tel: (613) 545-2034   Email: Don.Coleman AT QueensU DOT CA